基于改进YOLOv5的军事目标识别方法
A Military Target Detection Method Based on Improved YOLOv5
万晓刚 1王伟1
作者信息
- 1. 西安工程大学计算机科学学院 西安 710600
- 折叠
摘要
针对战场环境下因背景干扰和军事目标尺度较小等原因导致误检、漏检的问题,提出一种基于改进YOLOv5的军事目标识别方法CB-YOLOv5.利用坐标注意力机制重构特征提取主干网络,增强网络对复杂背景下军事目标的特征提取能力;在特征融合网络中引入BiFPN,减少浅层特征信息的丢失,提高对弱小目标的检测能力.在自建数据集下实验表明,改进后算法mAP达到93.8%,比原模型提升了3.5%,可以有效识别战场环境下多尺度军事目标.
Abstract
To address the problem of false detection and missed detection due to background interference and small scale of military targets in battlefield environment,a military target recognition method CB-YOLOv5 based on improved YOLOv5 is pro-posed.The feature extraction backbone network is reconstructed by using coordinate attention mechanism to enhance the feature ex-traction capability of the network for military targets in complex background.BiFPN is introduced in the feature fusion network to re-duce the loss of shallow feature information and improve the weak targets can be detected.Experiments under the self-built dataset show that the improved algorithm mAP reaches 93.8%,which is 3.5%better than the original model,and can effectively identify multi-scale military targets in the battlefield environment.
关键词
目标识别/YOLOv5/注意力机制/特征融合Key words
target detection/YOLOv5/attention mechanism/feature fusion引用本文复制引用
基金项目
2021年中国高校产学研创新基金项目(2021ALA02002)
出版年
2024
国家科技期刊平台
NSTL
万方数据